Sankofa/scripts/check-cluster-status.sh

#!/bin/bash
# check-cluster-status.sh
# Checks the status of a Proxmox cluster

set -euo pipefail

# Load environment variables
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
if [ -f "${SCRIPT_DIR}/../.env" ]; then
    set -a
    source <(grep -v '^#' "${SCRIPT_DIR}/../.env" | grep -v '^$' | sed 's/^/export /')
    set +a
fi

# Colors
GREEN='\033[0;32m'
RED='\033[0;31m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
CYAN='\033[0;36m'
NC='\033[0m'

# Configuration
CLUSTER_NAME="${1:-sankofa-sfv-01}"
NODE1_IP="192.168.11.10"
NODE1_NAME="ML110-01"
NODE1_TOKEN="${PROXMOX_TOKEN_ML110_01:-}"
NODE2_IP="192.168.11.11"
NODE2_NAME="R630-01"
NODE2_TOKEN="${PROXMOX_TOKEN_R630_01:-}"

log() {
    echo -e "${GREEN}[INFO]${NC} $1"
}

error() {
    echo -e "${RED}[ERROR]${NC} $1" >&2
}

warn() {
    echo -e "${YELLOW}[WARN]${NC} $1"
}

info() {
    echo -e "${BLUE}[INFO]${NC} $1"
}

check_cluster_status() {
    local endpoint=$1
    local token=$2
    local node_name=$3
    
    echo ""
    info "=== ${node_name} Cluster Status ==="
    
    # Check cluster status
    local status_response=$(curl -k -s -H "Authorization: PVEAPIToken ${token}" \
        "${endpoint}/api2/json/cluster/status" 2>/dev/null)
    
    if echo "$status_response" | jq -e '.data' >/dev/null 2>&1; then
        local cluster_name=$(echo "$status_response" | jq -r '.data[0].name // "unknown"')
        local node_count=$(echo "$status_response" | jq -r '.data | length')
        
        if [ "$cluster_name" != "null" ] && [ -n "$cluster_name" ]; then
            log "✓ Cluster found: ${cluster_name}"
            log "  Node count: ${node_count}"
            echo "$status_response" | jq -r '.data[] | "  • \(.name) - Type: \(.type) - Status: \(.status // "unknown")"'
        else
            warn "Cluster status endpoint accessible but no cluster data"
        fi
    else
        local error_msg=$(echo "$status_response" | jq -r '.message // "Unknown error"')
        if echo "$error_msg" | grep -q "Permission check failed"; then
            warn "Permission denied (may need Sys.Audit permission)"
        else
            warn "Not in cluster or cluster not accessible: ${error_msg}"
        fi
    fi
}

check_cluster_nodes() {
    local endpoint=$1
    local token=$2
    local node_name=$3
    
    echo ""
    info "=== ${node_name} Cluster Nodes ==="
    
    local nodes_response=$(curl -k -s -H "Authorization: PVEAPIToken ${token}" \
        "${endpoint}/api2/json/cluster/config/nodes" 2>/dev/null)
    
    if echo "$nodes_response" | jq -e '.data' >/dev/null 2>&1; then
        local node_count=$(echo "$nodes_response" | jq -r '.data | length')
        
        if [ "$node_count" -gt 0 ]; then
            log "✓ Cluster nodes found: ${node_count}"
            echo "$nodes_response" | jq -r '.data[] | "  • Node: \(.node) - Node ID: \(.nodeid) - Votes: \(.votes)"'
        else
            warn "No nodes found in cluster configuration"
        fi
    else
        local error_msg=$(echo "$nodes_response" | jq -r '.message // "Unknown error"')
        if echo "$error_msg" | grep -q "Permission check failed"; then
            warn "Permission denied (may need Sys.Audit permission)"
        else
            warn "Cluster nodes not accessible: ${error_msg}"
        fi
    fi
}

check_cluster_config() {
    local endpoint=$1
    local token=$2
    local node_name=$3
    
    echo ""
    info "=== ${node_name} Cluster Configuration ==="
    
    local config_response=$(curl -k -s -H "Authorization: PVEAPIToken ${token}" \
        "${endpoint}/api2/json/cluster/config" 2>/dev/null)
    
    if echo "$config_response" | jq -e '.data' >/dev/null 2>&1; then
        local cluster_name=$(echo "$config_response" | jq -r '.data.clustername // "unknown"')
        
        if [ "$cluster_name" != "null" ] && [ -n "$cluster_name" ]; then
            log "✓ Cluster name: ${cluster_name}"
            
            if [ "$cluster_name" = "$CLUSTER_NAME" ]; then
                log "  ✓ Matches expected cluster name: ${CLUSTER_NAME}"
            else
                warn "  ⚠ Cluster name mismatch. Expected: ${CLUSTER_NAME}, Found: ${cluster_name}"
            fi
        else
            warn "Cluster config accessible but no cluster name found"
        fi
    else
        local error_msg=$(echo "$config_response" | jq -r '.message // "Unknown error"')
        if echo "$error_msg" | grep -q "Permission check failed"; then
            warn "Permission denied (may need Sys.Audit permission)"
        else
            warn "Cluster config not accessible: ${error_msg}"
        fi
    fi
}

main() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════════╗"
    echo "║          Proxmox Cluster Status Check                        ║"
    echo "╚══════════════════════════════════════════════════════════════╝"
    echo ""
    
    info "Checking cluster: ${CLUSTER_NAME}"
    echo ""
    
    # Check Node 1
    check_cluster_status "https://${NODE1_IP}:8006" "${NODE1_TOKEN}" "${NODE1_NAME}"
    check_cluster_nodes "https://${NODE1_IP}:8006" "${NODE1_TOKEN}" "${NODE1_NAME}"
    check_cluster_config "https://${NODE1_IP}:8006" "${NODE1_TOKEN}" "${NODE1_NAME}"
    
    # Check Node 2
    check_cluster_status "https://${NODE2_IP}:8006" "${NODE2_TOKEN}" "${NODE2_NAME}"
    check_cluster_nodes "https://${NODE2_IP}:8006" "${NODE2_TOKEN}" "${NODE2_NAME}"
    check_cluster_config "https://${NODE2_IP}:8006" "${NODE2_TOKEN}" "${NODE2_NAME}"
    
    echo ""
    echo "╔══════════════════════════════════════════════════════════════╗"
    echo "║                    Summary                                   ║"
    echo "╚══════════════════════════════════════════════════════════════╝"
    echo ""
    
    # Try to determine overall status
    local node1_status=$(curl -k -s -H "Authorization: PVEAPIToken ${NODE1_TOKEN}" \
        "https://${NODE1_IP}:8006/api2/json/cluster/status" 2>/dev/null | \
        jq -r 'if .data then "in-cluster" else "standalone" end' 2>/dev/null || echo "unknown")
    
    local node2_status=$(curl -k -s -H "Authorization: PVEAPIToken ${NODE2_TOKEN}" \
        "https://${NODE2_IP}:8006/api2/json/cluster/status" 2>/dev/null | \
        jq -r 'if .data then "in-cluster" else "standalone" end' 2>/dev/null || echo "unknown")
    
    if [ "$node1_status" = "in-cluster" ] && [ "$node2_status" = "in-cluster" ]; then
        log "✓ Both nodes appear to be in a cluster"
    elif [ "$node1_status" = "standalone" ] && [ "$node2_status" = "standalone" ]; then
        warn "Both nodes are standalone (not clustered)"
    else
        warn "Mixed status: Node 1: ${node1_status}, Node 2: ${node2_status}"
    fi
    
    echo ""
    info "Note: Some checks may require additional API permissions (Sys.Audit)"
    info "For full cluster status, use Proxmox web UI or SSH: pvecm status"
    echo ""
}

main "$@"